Affiliation:
Distinguished University Professor, Mechanical Engineering, Maryland Energy Innovation Institute, University of Maryland
Title: High Intensity Distributed Combustion for Near Zero Emission and High Performance
Abstract: Clean fuel burning using a wide variety of fuels is extremely important for good human health and the environment. Colorless Distributed Combustion (called CDC) is a novel method to enhance flame stability, thermal field uniformity, combustion efficiency and significantly reduce pollutants emission, including noise and instability. The CDC is achieved through the use of a carefully prepared fuel-oxidizer mixture along with reactive species generated from within the combustor volume.
Bio: Ashwani Gupta is Distinguished University Professor at the University of Maryland. He received his PhD from the University of Sheffield, UK. He was awarded a Higher Doctorate (DSc) from the University of Sheffield, and also from the University of Southampton, UK. He received Honorary Doctorates from the University of Wisconsin Milwaukee, King Mungkut University of Technology, North Bangkok (bestowed by the Princess of Thailand), and the University of Derby, UK. His main research interests have been in the fields of Combustion, Air pollution, Propulsion, High temperature air combustion, Swirl flows, Diagnostics, Fuel sprays, Fuel reforming, Sensors, Micro-scale combustion, and Wastes to clean energy conversion. He has co-authored three books, published 290+ journal papers, 540+ conference papers, 22 book chapters, and 19 edited books. He is an Honorary Fellow of ASME and RAeS (UK), and a Fellow of AIAA, SAE and AAAS, and Member of the European Academy of Sciences and Arts (EASA).
Affiliation:
Director, Hypersonic Vehicle Simulation Institute at United States Air Force Academy
Title: Simulating Hypersonic Heat: How Can We Improve our Models and Methods?
Abstract: Hypersonic vehicles flying at Mach numbers above five can experience temperatures in excess of several thousand Kelvin. Dealing with the heating caused by these high temperatures can require either passive or active cooling systems, but these thermal protection systems cannot be designed well without accurate simulations or predictions of the heat loads (both instantaneous and integrated over time). Understanding where the high temperatures come from, how they flow around a vehicle, and how they interact with the surface and structure of the vehicle is crucial to the success of the design. Current capabilities in predicting hypersonic heating will be examined, including the impact of current turbulence and transition models on those predictions, and approaches to improving our predictive capabilities will be explored.
Bio: Dr. Cummings is Professor of Aeronautics and Managing Director of the DoD HPCMP Hypersonic Vehicle Simulation Institute at the US Air Force Academy. From 2015-2018 he was the Technical Director at AFOSR's European Office of Aerospace Research & Development in London. Dr. Cummings is a graduate of the University of Southern California where he received his Ph.D. in Aerospace Engineering, and also has earned B.S., B.A., and M.S. degrees from California Polytechnic State University. He currently serves as deputy editor of the Journal of Spacecraft and Rockets, and is an associate editor of the Journal of Aircraft and Aerospace Science and Technology. He is co-author of the Sixth Edition of Aerodynamics for Engineers and lead author for Applied Computational Aerodynamics. Dr. Cummings has previously worked at Hughes Aircraft Company, NASA Ames Research Center, and California Polytechnic State University. He is a Fellow of the American Institute of Aeronautics and Astronautics and the Royal Aeronautical Society.
Affiliation:
President and Founding Partner, ACRi
Title: The Coming Revolution in CFD: Physics Informed Machine Learning
Abstract: Physics Informed Machine Learning (PIML) appears to be a rapidly developing technique with the potential to revolutionize the computing sciences in general and computational fluid dynamics in particular. The first known instance of using the neural network to solve partial differential equations, without using training data, was Lagaris (1998). They used a meshless collocation method and determined the unknown coefficients of the neural network by minimizing the residual of the governing equations at the collocation points.
The development of the physics informed neural network (PINN) by Raissi et.al.(2017) involved the joint use of data driven techniques and the governing equations and sparked widespread interest in this technique. It appears likely that the PINN may take its place alongside the popular numerical techniques like the finite difference, finite volume and the finite element methods in CFD and scientific computing. The neural network is a continuous and differentiable function that is valid over the whole domain, is quick to evaluate and is more compact in terms of storage than the results of a typical numerical method.
The neural network is a powerful function approximator (Hornik 1980). Thus, it can be used to fit data generated by numerical solutions of the governing partial differential equations and/or data generated from experimental / field measurements and by directly minimizing the residuals of the the governing equations. This residual minimization technique can also be used to calibrate unknown coefficients of the governing equations by using experimental / measured data. This method makes the PINN very useful in solving inverse problems, uncertainty quantification and more generally, renders the PINN useful in situations wherever fast surrogate models are desired. Neural networks can also be configured to simultaneously output a prediction confidence interval alongside the actual results, which is useful for uncertainty quantification. This makes the PINN a good choice for embedding inside digital twins (Greaves 2002).
However, in spite of all the above advantages, the PINN has its shortcomings. One difficulty is that the training process is slow and computationally intensive. The convergence of the training (residual minimization) process is slow and often sensitive to the architecture of the network. At present, the PINN only takes the spatial coordinates of the collocation points and time as inputs. This implies that the PINN is trained for a specific boundary and initial condition, which restricts its use in practical systems. Studies that try to generalize PINN to a range of boundary and initial conditions are a matter of active research.
The talk concludes with a case study to illustrate the use of the neural network and PINN in problems related to CFD applications.
Bio: Over 50 years of experience in Computational Fluid Dynamics (CFD) and numerical simulation of flow, heat and mass transport processes in engineering and environmental sciences. Key member of the 3-person team led by Prof. D. B. Spalding that invented the Finite Volume Method (FVM) for CFD in mid 1960's.
In 1979, established the ACRi group of companies (www.acricfd.com) that now has offices in Los Angeles (USA), Nice (France) and Bangalore (India). In 2011, founded a non-profit CFD Virtual Reality Institute (www.CFDVRi.org) to further the cause of CFD education, training and R&D.
Experience spans a range of problems including those related to design, production, operation and environmental impact of industrial and urban projects, specifically the analysis and management of natural and man-made disaster scenarios such as Tsunamis, industrial fires and explosions, and hazardous and nuclear releases in air and water.
Provided services to over 200 leading industrial, government and research organizations in over 25 countries, and consulted with many Fortune 500 corporations including ANDRA, Aerospatiale, Allied Signal, Allison Gas Turbines, ARAMCO, ARCO, BARC, Bechtel, Boeing, BP, BRNS, Brown Boveri, Chevron, Exxon, Fluor, FMC, GE, General Motors, GIE Hyperspace, GTRE, IBM, Idaho National Engineering Laboratory, NASA, National Academy of Sciences, Oak Ridge National Laboratory, Rockwell, Rolls-Royce, Sandia National Laboratory, Shell, SNECMA, Sohio, Sulzer, TOTAL, Westinghouse, URS, USAF and Woodward Clyde, Dames & Moore and Woodward Clyde.
Principal author of the ANSWER®, PORFLOW®, TIDAL® and RADM CFD software tools, which can deal with a broad spectrum of problems in fluid dynamics, heat and mass transport, and environmental pollution. These tools are widely employed by commercial, academic and research organizations.
Author or co-author for 10 books and over 200 technical publications, and regular reviewer for a number of technical journals. Delivered keynote and invited talks at more than 100 conferences and seminars.
Fellow of the ASME and served as Chairman of the IIT Kanpur Foundation Board. IIT Gandhinagar Advisory Board, and as an advisor to DeitY (Government of India) and the Indian Army.